Variant Effect Coding
Bioinformatics dataset of more than 50,000 annotated genetic mutations, making it possible to train models for the classification of pathogenic vs benign variants.
Description
Variant Effect Coding is a structured dataset for bioinformatics, containing more than 50,000 genetic variants. It comes from the fusion of data from ClinVar (pathogenic variants) and GnomAD (benign high-frequency variants). The data is organized by chromosomes and ready for supervised classification tasks.
What is this dataset for?
- Training supervised learning models for the detection of pathogenic genetic mutations
- Testing LLMs architectures applied to genomics
- Exploring biological multimodal reasoning approaches based on codons and annotations
Can it be enriched or improved?
Yes, this dataset can be enriched by adding other types of variants, functional annotations, or even by crossing with clinical data. It is also possible to add vector representations from models such as DNA-BERT or other organic LLMs.
🔎 In summary
🧠 Recommended for
- Genetic researchers
- Bioinformatics engineers
- AI/DNA projects
🔧 Compatible tools
- Hugging Face Datasets
- Scikit-learn
- PyTorch
- DNA-BERT
- BioBert
💡 Tip
To maximize performance, try special encoding of sequences before fine-tuning (e.g. k-mer tokenization).
Frequently Asked Questions
Does this dataset include complete DNA sequences?
No, these are precise genetic variants with annotation, not complete genome sequences.
Can this dataset be used to detect new mutations?
Not directly, but it is perfect for training a model that could then be generalized to new cases.
Is it suitable for training with LLMs specialized in biology?
Yes, it has been used in recent research on training multimodal DNA-LLMs.
